MCP & agentic exploration

Rankless wraps its low-latency citation backend in the Model Context Protocol, so any MCP client can explore the data. On top of it, an LLM agent mines interesting stories — and every number it publishes is re-issued from the backend, never taken from the model.

Exploration sessions

Each session's command, metadata, and verified outputs. Every number was re-issued from the backend, not written by the model.

No public sessions yet.

Connect your agent

Point any MCP client at the hosted endpoint (streamable-http):

Claude Code

claude mcp add --transport http rankless https://alpha-api.rankless.org/mcp

MCP config (.mcp.json / Cursor / Desktop)

{
  "mcpServers": {
    "rankless": {
      "type": "http",
      "url": "https://alpha-api.rankless.org/mcp"
    }
  }
}

Or run the stdio proxy against the public REST API

RANKLESS_BE_URL=https://alpha-api.rankless.org/v1 uv run -m mcp_server

Tools

Each tool proxies one backend endpoint and returns rankless_url backlinks; ids must come from the resolution tools, never guessed.

search_entities /v1/names/{entity_type}

Search entities by name; the ONLY legitimate way to turn a name into ids.
get_top_entities /v1/tops

Top entities per type (institutions, authors, sources, countries, subfields).
get_entity_profile /v1/views/{entity_type}/{semantic_id}

Full profile of one entity: totals, yearly series, top relations, similars.
get_entity_stats /v1/stats/{entity_type}/{semantic_id}

Lifetime + year-windowed paper/citation counts, top citing subfields.
get_citation_tree /v1/trees/{entity_type}/{semantic_id}

Hierarchical citation-impact breakdown of an entity, flattened to top-N.
get_papers /v1/works/{entity_type}/{semantic_id}/{offset}

Papers of an entity. sort="citations" ranks by citation count first.
get_peers /v1/peers/{entity_type}/{semantic_id}

Peer entities (comparable size + field profile) and top subfields.
lookup_orcid /v1/orcid/{orcid}

Resolve an ORCID iD (e.g. 0000-0001-7896-6217) to a rankless author.

Exploration foci

A session is scoped to any of these; the agent separates its findings accordingly.

share: genuinely interesting, TRUE, shareable findings: surprising rankings, striking trends, unexpected cross-field impact, David-vs-Goliath comparisons, human-interest angles. Classify each with `share_kind`: entity-value (spotlight one entity's standing), comparison (two+ entities), strengths-weaknesses (where an entity dominates vs lags), analysis (a deeper multi-number read), or other.
query: the result of a specific investigation. Report the answer plainly in `description`, set `question` to the exact question answered, and back it with metrics.
data-issue: possible data problems: implausible counts, duplicates, zeros, extreme outliers, wrong attributions, garbled/mojibake names, mismatched field/journal mappings. Set `issue_kind`: "ledger-fix" if a single logged-in user edit would correct it - then fill `ledger_suggestion` (kind one of merge_authors / merge_papers / claim_paper / disown_paper / add_paper_request, a human `note`, and any ids/names you can infer in `details`; omit ids you cannot see) - or "investigation" if it needs more digging first.

Scoping a session

--backend: one of ['local', 'live'] or a full /v1 base URL (default: local).
--foci: comma list of ['share', 'query', 'data-issue'] or 'all' (default: query when --investigate/--question, else share).
--subject: center the round on one entity/scope, e.g. 'Hungary' or 'authors:balazs-lengyel'.
--question: a specific investigation for the query focus.
--investigate: deepen a past finding: '<run>' or '<run>:<id>' under the writeups dir.

Resources & prompts

rankless://schema/entity-types: # Rankless entity types
rankless://guide/agent: # Using the rankless tools
author_impact_report prompt: Structured research-impact report for one author.